Publish or Perish culture + ChatGPT = Rampant academic fraud in the coming years. I guess the real-world productivity of scientists (not just paper-piling productivity) will take a large hit, as they are fed false data and lose a lot of time trying to replicate bogus findings and sifting through all those spam papers to find the good ones.
When we start being getting technical and original, as research should be, ChatGPT fails completely. I have read some AI-generated attempts at imitating actual research and it becomes extremely obvious after the first paragraph.
The result looks a bit like the kind of pseudoscientific bullshit used by snake oil merchants: the words are here, the writing is fine, but it is nonsense. It may be good enough for people who lack proper scientific education, but I don't think it will last more than a few minutes in the hands of a scientific reviewer.
> I have read some AI-generated attempts at imitating actual research
For AI to actually write up research, it would first need the tools to actually do research (ignoring the cognitive capacity requirements that everyone focuses on.)
Why do you think ChatGPT plays a major role in increasing fraud? ChatGPT doesn't seem necessary to make up data believable data - maybe even the opposite. Maybe it makes writing the paper easier, but I don't think that will have a huge impact in scientific fraud.
People don't like to lie, so the more they have to lie to commit fraud the fewer will commit fraud. If they have to lie up a whole paper very few will do it, if they just have to click a button and then the only lie is to say they did it on their own then many more will do it.
as a plausible example I have experienced when attempting to use it for writing papers:
I give it a list of steps I did to generate some data -
it writes a long winded explanation of how to set it up that is similar but subtly different, which would lead to the results being dramatically different. The worst part is because of the nature of how these things work, the resultant steps is closer to how one might expect the solution to work.
This, if published, could result in hundreds of lost hours for someone else trying to implement my successful solution the wrong way
